←back to thread

Show HN: Semlib – Semantic Data Processing

(github.com)
59 points anishathalye | 1 comments | 15 Sep 25 13:45 UTC | HN request time: 0s | source
Show context
Y_Y ◴[15 Sep 25 14:34 UTC] No.45250210[source]▶

  >>> await sort(presidents, by="right-leaning")
  ['Jimmy Carter', 'Bill Clinton', 'George H. W. Bush', 'Ronald Reagan']
Is this supposed to be impressive? GIGO, if you want to vibe-classify your data then go right ahead, but I hope nobody serious relies on it.
replies(3): >>45250378 #>>45250441 #>>45251590 #
1. hobofan ◴[15 Sep 25 14:49 UTC] No.45250378[source]▶
Why not?

List-sorting/prioritizing list is among one of the best use-cases for LLMs, especially if the metrics for it are fuzzy, e.g. "what are the 10 sales lead of this list of 1000 that I should prioritize".

One of the more interesting approaches for that is arbitron[0], which does pairwise ranking with multiple metrics/agents to provide a multi-faceted sorting.

[0]: https://github.com/davidgasquez/arbitron